CEE02- Generating High Resolution Rainfall Data Using Statistical - - PowerPoint PPT Presentation

CEE02- Generating High Resolution Rainfall Data Using Statistical Techniques

By: Foo Xiang Hua Matthew, St. Andrews Junior College Under the guidance of: Associate Professor QIN Xiaosheng, School of Civil and Environmental Engineering

General Circulation Models (GCMs)

https://drtimball.ca/2012/static-climate-models-in-a-virtually-unknown-dynamic-atmosphere/

• General Circulation Models (GCMs)

○ Surface pressure ○ Temperature ○ Relative humidity ○ Geopotential height (gravity- adjusted height) ○ Calculated airflow variables

General Circulation Models (GCMs)

Approximately 300*300 KM at Equator

General Circulation Models (GCMs)

Approximately 300*300 KM at Equator

Downscaling

Downscaling: Improve resolution/ reduce coarseness Dynamical Embed a Regional Climate Model (RCM) Accounts for site- specific physical conditions Statistical Downscale the GCM data Identifes correlations between GCM data and observed data

https://www.climateprediction.net/climate-science/climate-modelling/regional-models/

Downscaling

• Advantages

○ Computationally inexpensive ○ Multiple scenarios, useful for uncertainty analysis

• Disadvantages

○ Assumes a significant and stationary relationship ○ Dependent on quality of GCM ○ Depends on selection of predictor variables

Downscaling: Improve resolution/ reduce coarseness Dynamical Embed a Regional Climate Model (RCM) Account for site- specific physical conditions Statistical Mathematical model Identify correlations between GCM data and observed data

• Advantages

○ Higher resolution of small scale atmospheric features

• Disadvantages

○ Computationally expensive ○ Dependent on quality of GCM ○ Sensitive to choice

• f boundary

conditions

Downscaling

https://www.climateprediction.net/climate-science/climate-modelling/regional-models/

Mathematical relationship

Statistical DownScaling Model (SDSM)

• Aim: Downscale precipitation over Singapore
• Study area: Singapore
• GCMs used (predictors):

○ Hadley Centre Coupled Model (HadCM3) ○ Canadian Earth System Model (CanESM2) ○ NCEP/NCAR Reanalysis data

• Rainfall records (predictand): Changi Climate Station daily precipitation
• Time period: 1967-2005 (hindcast)
• Software used: Statistical Downscaling Model (SDSM)

Statistical DownScaling Model (SDSM)-- Methodology

• Step 1- Data transformation
• Four experiments were conducted
• Two experiments involved

predictand transformation to 0.25 power

Experiments HadCM3 Non transformed predictand Predictand transformed to power 0.25 CIMP5 Non transformed predictand Predictand transformed to power 0.25

Statistical DownScaling Model (SDSM)-- Methodology

• Step 2- Screen Variables
• Conditional process for

precipitation, unconditional for rainfall

• Test predictors in correlation matrix
• Select predictors with highest

correlations

Statistical DownScaling Model (SDSM)-- Methodology

• Step 3- Calibrate Model
• Predictors with highest correlation

used to calibrate model

• .PAR file created

Statistical DownScaling Model (SDSM)-- Methodology

• Step 4- Weather Generator
• .PAR file used as input
• .SIM and .OUT file created
• Number of ensembles = Number of

generated scenarios

Results

• Analysed in

○ Monthly series ○ Monsoon seasons ○ Weekly analysis

• Analysed features

○ Mean ○ Variance ○ 90th percentile ○ Maximum ○ Minimum ○ Percentage of wet days ○ Percentage of dry days

0.000 2.000 4.000 6.000 8.000 10.000 12.000 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 Rainfall in mm Year July rainfall mean CanESM2 July mean CanESM2 transforme d mean 0.000 2.000 4.000 6.000 8.000 10.000 12.000 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 Rainfall in mm Year July rainfall mean HadCM3 rainfall mean HadCM3 transform ed rainfall mean

Results-- Monthly analysis

• Each July from 1967

to 2005 was analysed for mean values

• Downscaled scenario

was unable to replicate variability

• HadCM3 was able to

reflect rate of rainfall increase (mm/ year) more accurately

CanESM2 HadCM3 Changi Transformed

• 0.009

0.018 0.041 Non- transformed

• 0.017

0.022

Results-- Monthly analysis

5 10 15 20 25 30 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 Rainfall in mm Year HadCM3 mean July rainfall mean 5 10 15 20 25 30 35 Rainfall in mm Year HadCM3 transforme d mean July rainfall mean

• Data transformation

appears to cause uncertainty to be more varied (height of error bars)

Results-- Monsoon analysis

• Each North East Monsoon dry season (January to early March) from 1967 to 2005 was analysed for maximums

0.000 50.000 100.000 150.000 200.000 250.000 300.000 Rainfall in mm Year NEM dry max rainfall CanESM2 transformed max HadCM3 transformed max CanESM2 max HadCM3 max

Results-- Weekly analysis

• Analysed for percentage
• f wet days in a week
• Five day week
• Year 1967
• Range was

underestimated

• Data transformation

leads to consistently higher values

• No distinct wet or dry

downscaled periods

0.000% 20.000% 40.000% 60.000% 80.000% 100.000% 120.000% 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 Percentage of wet days Week number Changi climate station week 1-72 HadCM3 week 1-72 HadCM3 transformed week 1-72 0.000% 20.000% 40.000% 60.000% 80.000% 100.000% 120.000% 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 Percentage of wet days Week number Changi climate station week 1-72 CanESM2 week1-72 CanESM2 transforme d 1-72

Conclusion

• Poorly generated scenarios due to

○ Missing December 1969 and October and November in 1971 in Changi Climate records ○ Large number of HadCM3 predictors registered missing values ○ Poor correlations between predictors and predictand

• Possible improvements

○ Use of bias correction ○ Alternative predictor selection methods– consider partial correlations

Conclusion

Thank you for your kind attention